Buffered and Registered Memory: Is it worth the expense?

[Matt]

Hey guys.

As the subject says, I'm pricing up some RAM. I'm looking for 2GB of
PC2-8500 and I'm coming across modules that are either "buffered" or
"unbuffered" and "registered" or "un-registered".

I've found a website that explains what these mean, and it seems to
emphasise that buffered and registered memory is only necessary in:

"servers and other mission-critical systems where it is extremely
important that the data is properly handled."

For an average consumer such as myself, should I save some pennies and
get unbuffered and unregistered RAM, or is their a genuine benefit to me?

Kind Regards,

Matt

 

[Frank McCoy]

For an average consumer such as myself, should I save some pennies and
get unbuffered and unregistered RAM, or is their a genuine benefit to me?

*IF* your motherboard supports buffered and registered ECC RAM, then get
ECC RAM. If not, then don't bother. Most don't (but should). ;-{

 

[Paul]

Hey guys.

As the subject says, I'm pricing up some RAM. I'm looking for 2GB of
PC2-8500 and I'm coming across modules that are either "buffered" or
"unbuffered" and "registered" or "un-registered".

I've found a website that explains what these mean, and it seems to
emphasise that buffered and registered memory is only necessary in:

"servers and other mission-critical systems where it is extremely
important that the data is properly handled."

For an average consumer such as myself, should I save some pennies and
get unbuffered and unregistered RAM, or is their a genuine benefit to me?

Kind Regards,

Matt

Registered memory, adds a register chip between the memory chips on the
DIMM, and the control/address bus. The purpose of doing that, is to reduce
the electrical load on the control/address bus.

By doing that, you can put more DIMMs on the same bus. That is handy for
building server boards, as they tend to have more DIMM slots than desktop
boards.

Unbuffered memory doesn't have the register chip in the path.

The other feature involves the data bit organization.

Memory equipped with check bits, allows errors to be detected or to be
corrected. Examples of methods are simple parity and ECC.

Not all chipsets have support for ECC. The X38 has support, but we're not
sure at this point, that it actually works. (The first report I've seen
so far, showed trouble when ECC DIMMs were used on an X38.)

If a user selects ECC equipped DDR2, at the moment that means using a
slower grade of module. Perhaps DDR2-667 would be the speed you'd expect
from an ECC equipped product - this is mainly because there is no interest
in the enthusiast market, for using ECC DIMMs.

The Athlon64 has the memory controller inside the processor itself, and
the chip supports a grand total of 144 bits of data. That is enough for
two 72 bit wide DIMMs. One of the options offered, is a protection method
called "Chip Kill", which has enough redundancy so that if a x4 wide memory
chip died, the memory could still function.

So a chip like the Athlon64 ensures that AMD users have an opportunity with
any motherboard, to have support for the extra check bits on an ECC module.
On Intel, the chipset must be specifically designed with the extra bits
added. And the practice in recent generations, has been for the most part
to not include the extra bits, on chipsets intended for Intel processors.
My 875P based board (DDR era) has ECC capability, but a lot of stuff more
modern than that does not. The X38 only supports ECC on DDR2, and the ECC
feature is missing if you buy an X38 with DDR3. It isn't even clear to me
yet, whether DDR3 has room for an ECC chip on it or not - I haven't seen a
DDR3 module with ECC included. (And navigating the JEDEC site is no fun, which
is why I haven't investigated further.)

While there is certainly value in having the ability to check the integrity
of data stored in the RAM, the industry is not making it easy for a desktop
user to get that feature. You have to decide what that is worth, as it affects
a few buying decisions.

Paul

 

[John Weiss]

As the subject says, I'm pricing up some RAM. I'm looking for 2GB of PC2-8500
and I'm coming across modules that are either "buffered" or "unbuffered" and
"registered" or "un-registered".

For an average consumer such as myself, should I save some pennies and get
unbuffered and unregistered RAM, or is their a genuine benefit to me?

Get what the motherboard supports. Most consumer-grade MoBos support
unregistered RAM. Opteron MoBos (especially multiple-socket types) normally
support buffered/registered RAM. Many/most server MoBos support or require ECC
RAM.

Generally, if it is not required, buffered/registered or ECC RAM is not worth
the extra cost for consumer machines.

 

[CBFalconer]

.... snip ...

Generally, if it is not required, buffered/registered or ECC RAM
is not worth the extra cost for consumer machines.

This misinformation needs to be squelched. ALWAYS get ECC memory
if your system supports it, and try to buy systems that support it.

There is no checking mechananism other than ECC available for
memory, and random memory failures, such as caused by cosmic rays,
can be fatal. The fault may not show up for weeks or months (or
more) after occuring, and such things as memtestXX will show no
fault. By this time all your backups are fouled.

The extra cost is very small, about 1/8 to 1/5 of the cost of the
simpler, faulty memory.

 

[John Doe]

This misinformation needs to be squelched. ALWAYS get ECC memory
if your system supports it, and try to buy systems that support
it.

No surprise the original crosspost would open a can of worms.

I've been building my own PC for over 15 years. Been making backup
copies of the Windows partition (the latter years including program
files) to the same hard drive for almost as long, that requires a
whole lot of data copying. I've never used ECC memory and I've never
had a memory error that destroyed data. Besides, I keep backup
copies of all important files from my hard drive. People who don't
put copies of important files on removable media are probably more
likely to lose data from a hard drive failure than from a non-ECC
memory failure.

There is no checking mechananism other than ECC available for
memory, and random memory failures, such as caused by cosmic rays,
can be fatal. The fault may not show up for weeks or months (or
more) after occuring, and such things as memtestXX will show no
fault. By this time all your backups are fouled.

The extra cost is very small, about 1/8 to 1/5 of the cost of the
simpler, faulty memory.

--
Chuck F (cbfalconer at maineline dot net)
<http://cbfalconer.home.att.net>
Try the download section.



--




Path: newssvr25.news.prodigy.net!newsdbm05.news.prodigy.net!newsdst01.news.prodigy.net!prodigy.com!newscon04.news.prodigy.net!prodigy.net!newshub.sdsu.edu!nntpserver.com!zeus.nntpserver.com!10.1.1.41.MISMATCH!pfilter-v0.1!not-for-mail
Message-ID: <47831965.935C3A88 yahoo.com>
Date: Tue, 08 Jan 2008 01:34:13 -0500
From: CBFalconer <cbfalconer yahoo.com>
Reply-To: cbfalconer maineline.net
Organization: Ched Research http://cbfalconer.home.att.net
X-Mailer: Mozilla 4.75 [en] (Win98; U)
X-Accept-Language: en
MIME-Version: 1.0
Newsgroups: alt.comp.hardware.pc-homebuilt,alt.comp.hardware,alt.comp.hardware.homebuilt
Subject: Re: Buffered and Registered Memory: Is it worth the expense?
References: <flundj$9qn$1 heffalump.dur.ac.uk> <NfudnRq3ZMxmbx_anZ2dnUVZ_g-dnZ2d comcast.com>
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Lines: 25
NNTP-Posting-Date: 08 Jan 2008 06:48:51 GMT
X-Complaints-To: abuse teranews.com
Xref: prodigy.net alt.comp.hardware.pc-homebuilt:500932 alt.comp.hardware:343689 alt.comp.hardware.homebuilt:232240

 

[John Doe]

Hey guys.

As the subject says, I'm pricing up some RAM.

I've found a website that explains what these mean, and it seems
to emphasise that buffered and registered memory is only necessary
in:

"servers and other mission-critical systems where it is extremely
important that the data is properly handled."

That's because businesses suffer from downtime and (very recently
generated) data loss. You and I don't.

For an average consumer such as myself, should I save some pennies
and get unbuffered and unregistered RAM, or is their a genuine
benefit to me?

If you're just a gamer, there is no benefit to you. Somebody
suggesting that a gamer needs ECC memory has pretty much lost it
IMO.

If you have any important files on your hard drive, keep at least
two copies (one backup, and one backup of the backup... both copies
don't have to be perfectly up to date) on removable media. Nobody
ever comes here complaining that their non-ECC memory caused
important data corruption. However, we do occasionally get a very
sad story about a user losing important data because of hard drive
failure.

 

[Frank McCoy]

Nobody
ever comes here complaining that their non-ECC memory caused
important data corruption. However, we do occasionally get a very
sad story about a user losing important data because of hard drive
failure.

Sometimes memory-failures are thought to be hard-drive failures.
When bad data from memory is saved, what's the first suspect when you
retrieve it? You don't suspect the memory, even when it's at fault.

Drives however *have* error-correction; so if you get bad data back from
the drive, most likely it was *stored* there.

The worst thing about memory errors, is that you never know when they
happen without at least parity and preferably ECC memory. Memory errors
happen ALL THE TIME; without ever being suspected.

Those who say, "I've been running for years and never had a memory
error," just don't KNOW they've probably had several ... If not from bad
memory, then from cosmic rays or just the normal background radiation
everywhere from the inevitable radioactive material built into about
everything. Carbon 14 being one of the more obvious; but uranium also
gets into just about everything ... not to mention all the stray stuff
from atomic tests in the background.

It's true that *most* of those errors don't cause visible problems; but
that's the worst part of the problem: The errors are NOT visible. If
you have data in memory and it gets hit, then saved, you have no idea
until something important *depends* on that data. Most other problems
just cause glitches or at worst system crashes. Those are fairly easily
recovered from if you're not handling sensitive or important data.

Of course, almost all such glitches and crashes are blamed on the
Operating System, especially if it's from MicroSoft.

There's little so frustrating as being in the middle of typing in
something important for HOURS and then having the system freeze just
when you're about to save it ... The sort of thing most likely to happen
with a flipped bit in your core programs.

However, it's quite true that if you're mainly a gamer and your
conversations aren't saved or important to you, then paying the extra
for ECC memory is probably wasted dough; as you'd be happier spending
the same money for a faster board or more memory.

The real problem is: Not all that many boards are offered with the
extras many people like me, that actually *include* ECC as an option.

 

[Al Dykes]

Sometimes memory-failures are thought to be hard-drive failures.
When bad data from memory is saved, what's the first suspect when you
retrieve it? You don't suspect the memory, even when it's at fault.

Drives however *have* error-correction; so if you get bad data back from
the drive, most likely it was *stored* there.

The worst thing about memory errors, is that you never know when they
happen without at least parity and preferably ECC memory. Memory errors
happen ALL THE TIME; without ever being suspected.

That doesn't pass the smell test. Lots of us run servers with GB of
ram chips that do have ECC and produce event logs for any error like
this even when it is "soft", i.e. it recovered.

I rarely see memory errors, recoverable or otherwise on machines that
log everything. I've run big computers with solid-state ECC memory
and hardware service logs since about 1982.

As an old phart, I too am amazed that we operate desktop machines
without ECC. It isn't just the ram chips, it's every bus in the
computer, too. Real world experience with MACs and x86 PCs that don't
produce mystery spreadsheet results or trash file systems due to a
flipped bit, corrupted huge business databases or random OS crashes
(with modern operating systems) says that undetected memory errors are
rare to non-existent.

My recollection is that memery errors were a problem circa 1975 and
cosmic rays were suspected. Someone discovered that the epoxy used to
package the memory was a source of alpha rays and they went away when
it was changed.

But it's been a long time and *my* memery is subject to parity errors
by now.

 

[John Doe]

In alt.comp.hardware.pc-homebuilt John Doe


Sometimes memory-failures are thought to be hard-drive failures.

Could be, but I'm talking about hard drive failures, the kind of
failure we hear about.

Those who say, "I've been running for years and never had a memory
error," just don't KNOW they've probably had several ... If not
from bad memory, then from cosmic rays or just the normal
background radiation everywhere from the inevitable radioactive
material built into about everything. Carbon 14 being one of the
more obvious; but uranium also gets into just about everything ...
not to mention all the stray stuff from atomic tests in the
background.

You know that "tinfoil hat" gibe? The guy it's based on was an
important scientist.

Of course, almost all such glitches and crashes are blamed on the
Operating System, especially if it's from MicroSoft.

One glaring contradiction is how much better Windows XP is over
prior consumer versions of Windows. My XP system runs for days
without having to be restarted. With prior versions, I sometimes had
to restart several times per day. In other words, if it wasn't
Windows, how come Windows XP is so much less unstable? If you think
Windows XP handles errors better, of course it does, the operating
system is supposed to handle errors. Prior to Windows XP, Windows
memory management sucked.

There's little so frustrating as being in the middle of typing in
something important for HOURS and then having the system freeze
just when you're about to save it ...

Maybe for Ernest Hemingway using pre Windows XP. But most programs
that handle important data allow automatic saving of data every X
number of minutes anyway.

Are you using pre Windows XP?

The sort of thing most likely to happen with a flipped bit in your
core programs.

However, it's quite true that if you're mainly a gamer and your
conversations aren't saved or important to you,

....or your system is reasonably stable

Not having copies of (any) important files is hugely more important
than memory type. If the hard drive fails, that can be a real
problem especially without backups.

If you can't afford to restart your computer or to lose recently
generated data, maybe that's justification for buying an ECC memory
system.

 

[Al Dykes]

Could be, but I'm talking about hard drive failures, the kind of
failure we hear about.


You know that "tinfoil hat" gibe? The guy it's based on was an
important scientist.


One glaring contradiction is how much better Windows XP is over
prior consumer versions of Windows. My XP system runs for days
without having to be restarted. With prior versions, I sometimes had
to restart several times per day. In other words, if it wasn't
Windows, how come Windows XP is so much less unstable? If you think
Windows XP handles errors better, of course it does, the operating
system is supposed to handle errors. Prior to Windows XP, Windows
memory management sucked.


Maybe for Ernest Hemingway using pre Windows XP. But most programs
that handle important data allow automatic saving of data every X
number of minutes anyway.

Any machine that has random undetected memory errors will eventually
corrupt the disk and if it hits an OS files like the registry, the OS
will corrupt, too.

 

[RobV]

*IF* your motherboard supports buffered and registered ECC RAM, then
get ECC RAM. If not, then don't bother. Most don't (but should).
;-{

Doesn't the cache on the CPU (L1 and L2) have ECC? At one time, at
least, I thought I read it did. However, just like Al Dykes, my wetware
is showing the problem of years of cosmic ( and other) rays over the
years.

 

[John Doe]

(e-mail address removed) (Al Dykes) wrote:

....

Any machine that has random undetected memory errors
will eventually corrupt the disk

You need to be more specific, for what it's worth in such a
mysterious context.

and if it hits an OS files like the registry, the OS will corrupt,
too.

Sounds like the bogeyman.

 

[John Doe]

Any machine that has random undetected memory errors will eventually
corrupt the disk and if it hits an OS files like the registry, the OS
will corrupt, too.

Al Dykes wrote:
"Real world experience with MACs and x86 PCs that don't produce
mystery spreadsheet results or trash file systems due to a flipped
bit, corrupted huge business databases or random OS crashes (with
modern operating systems) says that undetected memory errors are
rare to non-existent."

Maybe I'm losing something in the translation, but sounds like
you're on both sides of the issue.

 

[Al Dykes]

Al Dykes wrote:
"Real world experience with MACs and x86 PCs that don't produce
mystery spreadsheet results or trash file systems due to a flipped
bit, corrupted huge business databases or random OS crashes (with
modern operating systems) says that undetected memory errors are
rare to non-existent."

Maybe I'm losing something in the translation, but sounds like
you're on both sides of the issue.

In some ways I am, but I can look at a raft of machines that don't
have parity crank out calculations 24x7 without a random glitch.


My big machines have ECC and log everyting. I get hits but very rarely
and they likely result in a board being swapped out at the next chance
for service. Recoverable ECC hits *not* considered normal.

Running a business database on a machine without ECC and logging is
fooling because a bad number in memory might be written to it and we
wouldn't know for days, months, ever and the recovery from that error
might backrupt the company.

If I were running a computer that was used in a life-critical app, I'd
want parity. I wanter what the medical safety regs say.

If a machine can use ECC memory, that is kind of pointless if the OS
can't record the fact when ECC hardware is used to fix a bit. I've
been away from Intel servers for a while but it took mobo
vender-written driver and administrator code to capture that data and
report it.

I don't know that this comes with any current comodity mobo and if XP
has the capability to capture the data.

 

[CBFalconer]

Doesn't the cache on the CPU (L1 and L2) have ECC? At one time,
at least, I thought I read it did. However, just like Al Dykes,
my wetware is showing the problem of years of cosmic (and other)
rays over the years.

Far from guaranteed. Investigate the potential purchase carefully.

 

[John Doe]

Running a business database on a machine without ECC and logging
is fooling because a bad number in memory might be written to it
and we wouldn't know for days, months, ever and the recovery from
that error might backrupt the company.

I understand that whether you need it depends on the application.
But I would argue that having backup copies of important data comes
before needing more reliable RAM. And, judging by occasional
discussion by visitors to this group, apparently lots of typical
users don't even have copies of important files.

If I were running a computer that was used in a life-critical app,
I'd want parity. I wanter what the medical safety regs say.

Guessing. Probably beyond ECC/whatever. Probably more difficult to
obtain than military specification hardware. There are both types of
groups here on USENET, they might know.

 

[Al Dykes]

I understand that whether you need it depends on the application.
But I would argue that having backup copies of important data comes
before needing more reliable RAM. And, judging by occasional

If the backups are of corrupted files, you are f'cked and if it goes
months before it is detected, what are you going to tell the boss?
Forget about all the busines we did in that interval? Going backwards
and findng which backup predates the corruption is essentially
impossible

If a bit is flipped between the reading of a block of data and writing
it back, the database integrity mechanism won't detect it or give you
a neat way of recovering.

If that bit is in meta-data it can f'ck huge amounts of data. If you
are lucky, it will be bad enough to cause the database software to
crash on the spot, but it won't be pretty.

 

[Mike Walsh]

Most home PCs don't need registered / buffered memory. My main home PC has 2 DIMMs with 9 memory chips on each DIMM and it work fine. I have seen PCs at work with 4 DIMMs with 36 memory chips on each DIMM. It is unlikely that the memory bus would drive all these chips if they were not buffered.

 

[kony]

Doesn't the cache on the CPU (L1 and L2) have ECC? At one time, at
least, I thought I read it did. However, just like Al Dykes, my wetware
is showing the problem of years of cosmic ( and other) rays over the
years.

Yes they do, though one having ECC is not a substitute for
the other.